In 2010, on the message board website LessWrong, a user named Roko posted a philosophical thought experiment: What if sometime in the future there is an evil artificial intelligence that comes into existence and not only punishes people who don’t do its bidding, but also punishes people today for not doing what was necessary to bring about the AI into existence later? So an evil AI from the future could essentially blackmail people into bringing about its very existence? The founder of the site, also the founder of the Machine Intelligence Research Institute, Eliezer Yudkowsky, was very angry at Roko and said: “YOU DO NOT THINK IN SUFFICIENT DETAIL ABOUT SUPERINTELLIGENCES CONSIDERING WHETHER OR NOT TO BLACKMAIL YOU. THAT IS THE ONLY POSSIBLE THING WHICH GIVES THEM A MOTIVE TO FOLLOW THROUGH ON THE BLACKMAIL.”
This all became known as Roko’s Basilisk, spawning many Internet memes and possibly even the relationship between Elon Musk and Grimes. I swear I did not make all this up. The Internet is weird.
Today’s column from Joel Rollins reminded me of this bizarre episode from the Internet (and if it’s hard to wrap your brain around this concept, you’re really not alone). When I think of a malevolent AI, I now see it anthropomorphized as Clippy.